This page may contain stale information. Last updated: 2026-06-12
Overview
Recursive is an AI research company co-founded by richard-socher focused on automated AI research and recursively self-improving systems. On June 11, 2026, Recursive published early SOTA results from an automated research system that autonomously proposes, implements, tests, and validates ML improvements.
Automated Research System
The system runs many research threads over long horizons, combines promising branches, validates against reward hacks, and compounds discoveries across modeling, optimization, and systems layers.
June 2026 Benchmark Results
| Benchmark | Recursive | Previous Best | Metric |
|---|---|---|---|
| nanochat-autoresearch | 0.9109 | 0.9372 | Validation BPB |
| nanogpt-speedrun | 77.5s | 79.7s | Time to 3.28 val loss |
| sol-execbench | 0.754 | 0.699 | Mean SOL (235 kernels) |
Recursive open-sourced artifacts from these runs for independent verification (2026-06-11-recursive-automated-ai-research).
Related
- richard-socher
- andrej-karpathy
- automated-research
- recursive-self-improvement
- ai-scientist
- gpu-kernel-optimization
Sources
- 2026-06-11-recursive-automated-ai-research
- 2026-06-11-recursive-automated-ai-research-36kr
- 2026-06-11-recursive-automated-ai-research-digg
- 2026-06-11-recursive-automated-ai-research-cryptobriefing